Reproducing Imagenet Results

Pytorch documentation reports accuracies of several models on imagenet data. Please see the link.

Furthermore, they reference a training script ( Can one reproduce the reported imagenet results by using the referenced script? Or do I need some more pre-processing. For example Inception-v3 uses some form of colour augmentation and as far I see the reference script doesn’t contain colour augmentation (may be I am wrong, I am a newbee in pyTorch).